Sie sind auf Seite 1von 4

Predicting Students Academic Performance

Using Support Vector Machine


Iti Burman1, Subhranil Som2
1
Department of Information Technology, Amity University Uttar Pradesh, Noida, India
2
Amity University Uttar Pradesh, Noida, India
1
itiburman0017@gmail.com, 2ssom@amity.edu

Abstract: Education, more often known as learning, is a way SVM is a learning model, a supervised technique, used for
of exchanging knowledge with the perspective of betterment classification analysis of data. The algorithm works by finding
of individuals and progress of the nation as well. The hyperplane which aims to widen the boundaries between two
objective of this paper is to help students to improve their sets of classes. The goal of classification is to separate the data
performance with the use of applications of data mining. It points into separate classes. SVM views these points as an n-
makes use of psychological features of students. The paper dimensional vector and tries to separate them with (n-1)
uses multi classifier Support Vector Machine (SVM) to dimensional hyperplane. This mechanism is commonly known
classify the learners in the category of high, average and low as linear classifier. Various hyperplanes may exist to classify
according to their academic scores. It is carried out using the data, but it focuses on finding the one which provides
linear kernel and radial basis kernel. It is noted that RBF maximum separation.
produces better results as compared to the linear kernel.
Predicting the performance of students in advance can Section 2 provides an overview of related work and discusses
advantage both the institution and learner to take about various prediction techniques and their accuracy rate
measurable steps in order to enhance the learning process. applied in this area. Section 3 explains the proposed
methodology used for prediction in our study. Results of our
Keywords: Education, Data mining, SVM, Student study and their comparative analysis are given in Section 4. At
Performance, Prediction, Psychology the end, section 5 provides the conclusion.

I. INTRODUCTION II. LITERATURE REVIEW


Education plays a vital role in a person’s life. It helps in Prediction, an EDM technique, aims at envisioning the value
overall development of an individual. Other than well known of a variable called predicted variable from some known set of
learner pedagogue relation, there are several ways of learning values called predictors from the data. One of the advantages
including narrating, conference meet, guidance, and analysis; of prediction in EDM can be found in identifying the slow
which is done at all stages of life. Education act as an input in learners in the classroom, so that at an early stage steps can be
shaping the life of an individual. A student undergoes various taken to improve their performance. Various classification
exams at various stages of life. The measure of success in the techniques such as Naïve Bayes, Multilayer Perception,
exams especially in university entrance exam provides a REPTree, J48 and SMO have been applied to oversee the
measure of their successful career as shown in [6]. Data students scoring low as discussed in [10]. Prediction has
mining as a field in research tries to extract useful information played an important role in order to provide excellent learning
or patterns from the collected, processed and analyzed data. It to the learners. Hamsa et al. [11] had developed a prediction
finds its relevance in various streams. Educational data mining model to envision the scholarly performance of students based
is a process of inferring interesting patterns from educational on their scholastic parameters. The model uses two
data that come from educational institutions and web based classification techniques – Fuzzy genetic algorithm and
data; i.e. online and offline. It is an upcoming research area to Decision tree; which targets the undergraduate and graduate
facilitate better learning among pupil. An individuals’ students. The model can also be used by the college and
academic achievement is also based upon numerous non various reputed companies to select the students for
academic parameters. Analysis done on Indian case study [1] recruitment. Shahiri et al. [12] has discussed about different
using Logistic Regression showed that non academic correlates data mining approaches that can be applied to educational data
have an effect on CGPI of the learners. Psychology is an area based on different parameters. It covers K nearest neighbor,
that deals with behavior and mind. It captures the actions Support vector machine, Decision tree, Naïve bayes and
following state of mind according to the situations. In context Neural network for classification. Gray et al. [13] studied the
of education it can be utilized to enhance the learning process psychometric factors of students to predict their academic
by understanding the individuals’ behavior related to learning achievement and uses neural network and decision tree
and pedagogy. approach for prediction. The factors considered under study

978-1-5386-9346-9/19/$31.00 ©2019 IEEE


include personality, motivation and learning strategies; the NN The SVM algorithm works as follows:
algorithm predicts with 69% of accuracy and decision tree
predicts with 65% of accuracy. Mishra et al. [14] has worked i. Separable case is the one in which data can be perfectly
upon psychometric, extracurricular and soft skills parameters linearly separated. Here, infinite numbers of boundaries are
of learners and envisioned with accuracy of 88% by using possible and it selects the optimal hyperplane wherein the
decision tree algorithm. Sembiring et al. [15] also worked on boundary gives the maximum distance. Given a function:
psychometric factors as well and applied SVM on it. The
prediction accuracy was found to be 83%. f(y) = x.y + z, (1)

Richardson et al. [16] worked on psychological parameters for SVM divides the data points as:
learner’s scholastic performance. The parameters are divided
into five major categories – Personality, Motivation, Self f(y) > 0, iff y ‫ א‬X, and (2)
regulatory learning strategies, Learning approach and
f(y) <= 0, iff y ‫ א‬Z (3)
Psychosocial contextual influences. Our study considered these
parameters and collected the data with questionnaire prepared The distance between the observation and the hyperplane is
based on these psychometric parameters. It was found that given by |x.y + z|/||x||, and the margin is given as 2/||x||.
students’ scholarly achievement also depend on non
intellectual correlates than just academic correlates [17]. Also, ii. Non separable case, where data points overlap. To classify
Meta analysis showed the correlation in students’ percentage these data points, SVM performs restructuring of the data with
and non intellectual correlate. the use of transformation function represented as (Φ). It works
by mapping the scalar dot product of the data points to a higher
III. PROPOSED METHODOLOGY enough dimension where linear separation becomes possible.
The steps of proposed methodology are depicted in Figure 1.
B. Analysis
A. Support Vector Machine
An important and most common task of Machine learning,
SVM is a supervised learning technique that aims to classify Classification, can be performed with various data mining
the data. It makes use of hyperplane for dividing the dataset techniques. This paper focuses on classifying the students’ data
into classes with the gap as wide as possible known as margin. based on psychometric components into three classes: High,
It generates parallel lines for creating partitions. The margin is Average and Low. Our problem is a multi classifier problem. It
the maximum distance between the nearest data points of the makes use of Linear Kernel and Radial Basis Function Kernel.
classes. To reduce the generalization error, largest margin is The dataset consists of the records of the students collected
selected. using Questionnaire based on their psychological parameters
which covers Personality, Motivation, Psychosocial contextual
influences, Learning strategies, Approach to learning and
Socio economic status [17]. The dataset consists of one
thousand records based on 29 non intellectual constructs of
students. For classification, we divide the dataset as - 70% of
the data has been used for training the model and testing is
done on the rest 30% of the dataset.

The linear kernel can be defined as the inner dot product [a, b]
and an arbitrary constant c, which can be mathematically
represented as:

k(a, b) = aTb + c (4)

The radial basis function kernel (RBF) can be defined as:

K (x,x') = exp [(-||x-x'||2) ], (5)

where x and x' are two input feature vectors, ||x-x'||2 is the
square of Euclidian distance and is calculated as: 1/2 2. The
value of the RBF kernel can be used as likeness measure which
Fig. 1. Proposed model varies between 0 and 1; it decreases with distance.

757
IV. RESULTS AND DISCUSSIONS
The proposed model for predicting the performance of students
is assessed using sensitivity, specificity and accuracy. The
results of training data set using different kernels have been
shown in figure 2 and figure 3; and the results of testing data
are discussed in table 1 and table 2.

A. Sensitivity
It is a statistical metric of accomplishment that measures the
positive values (like in our study, correctly identifying the
students in the category of High, Average and Low according
to the given parameters).

(6)

B. Specificity

It measures the negative or false values.

(7)

C. Accuracy
It is the statistical biasness that measures the trueness i.e.
difference between the observed value and the true value.

TP  TN
Accuracy u 100
TP  TN  FP  FN (8)
Fig. 3. Results of Training data set using Radial Basis Kernel
where, TP represents no. of true positives, FP represents no. of
TABLE I: CONFUSION MATRIX
false positives, TN represents no. of true negatives and FN
represents no. of false negatives. Prediction Linear Kernel Radial Basis Function
Kernel
H L A H L A
H 100 0 52 140 0 14
L 0 6 0 0 6 0
A 53 0 88 13 0 126

TABLE II: OVERALL STATISTICS

Statistics Linear Kernel Radial Basis


Accuracy 0.6488 0.9097
95% CI 0.5918, 0.7029 0.8713, 0.9396
p-value 1.158e-06 2.2e-16
High Low Average High Low Average
Sensitivity 0.6536 1.0000 0.6286 0.9150 1.0000 0.9000
Specificity 0.6438 1.0000 0.6667 0.9041 1.0000 0.9182
Fig. 2. Results of Training data set using Linear Kernel

758
machine for classification of data and predict learners’ CGPI.
According to the statistics given in the Table 2 it was found
that Radial Basis Function kernel gives more accurate results
than Linear Kernel which is approximately 90%.

ACKNOWLEDGEMENT
I would like to express my sincere gratitude to Vivekananda
Institute of Professional Studies and Amity University Uttar
Pradesh for their continuous support.

REFERENCES
[1] Ranjan and R. Ranjan, “Modelling Key parameters in Higher
Education using Logistic Regression: an Indian case based Data
Analysis”, 4th International Conference on Reliability, Infocom
Technologies and Optimization (Trends and Future Directions),
pp. 365-369, 2015, IEEE, 2015.
[2] H. Goker, and H. I. Bulbul, “Improving an early warning
Fig. 4. SVM using Linear Kernel system to prediction of student examination achievement”, In
Machine Learning and Applications (ICMLA), 13th
The tuning of the model is done using grid with values ranging International Conference on pp. 568-573, IEEE, December
from 0 to 5. It was found that Linear kernel provides the best 2014.
values at tuning parameter C=2 as shown above in Figure 4; [3] P. Kaur, M. Singh, and G. S. Josan, “Classification and
when the model run using Radial Grid Kernel, it calculated prediction based data mining algorithms to predict slow learners
variations and provides the best results of sigma at 0.1 and C at in education sector”, Procedia Computer Science, 57, pp. 500-
2 as shown below in Figure 5. 508, 2015.
[4] H. Hamsa, S. Indiradevi, and J. J. Kizhakkethottam, “Student
academic performance prediction model using decision tree and
fuzzy genetic algorithm”, Procedia Technology, 25, pp. 326-
332, 2016.
[5] A. M. Shahiri, and W. Husain, “A review on predicting
student's performance using data mining techniques”, Procedia
Computer Science, 72, pp. 414-422, 2015.
[6] G. Gray, C. McGuinness and P. Owende, “An application of
classification models to predict learner progression in tertiary
education, in: Advance Computing Conference (IACC)”, IEEE
International, pp. 549–554, 2014.
[7] T. Mishra, D. Kumar and S. Gupta, “Mining students’ data for
prediction performance”, Proceedings of the 2014 Fourth
International Conference on Advanced Computing &
Communication Technologies, ACCT ’14, IEEE Computer
Society, Washington, DC, USA, 2014, pp. 255–262.
doi:10.1109/ACCT.2014.105.
[8] S. Sembiring, M. Zarlis, D. Hartama, S. Ramliana and E. Wani,
Fig. 5. SVM using Radial Grid Kernel “Prediction of student academic performance by an application
of data mining techniques”, International Conference on
V. CONCLUSION Management and Artificial Intelligence IPEDR, Vol. 6, pp.
110–114, 2011.
This paper focuses on non intellectual parameters of students
[9] M. Richardson, C. Abraham, and R. Bond, “Psychological
which affect their study and academic growth. Utilization of correlates of university students' academic performance: A
data mining in the field of education can prove to be a boon for systematic review and meta-analysis”, Psychological bulletin,
the society. Psychometric analyses of students’ behavior in 138(2), pp. 353, 2012.
respect of learning help in enhancing their academic [10] I. Burman, S. Som and S. A. Hossain, “Meta Analysis of
performance. Various mining techniques like neural network, Psychometric Measures and Prediction of Student’s Learning
decision tree, KNN, naïve bayes and SVM have been applied Behaviour using Regression Analysis and SVM”, Jour of Adv
to the educational data covering psychological factors. As Research in Dynamical and Control systems, Vol. 10, 02-
discussed in section 3, the accuracy rate of previous studies is Special issue, pp. 291-298, 2018.
less than 89%. Our proposed model uses Support vector

759

Das könnte Ihnen auch gefallen